Safety monitoring method for rehabilitation training of senile weak patient

Abstract

The invention relates to a safety monitoring method for rehabilitation training of an elderly weak patient. The safety monitoring method is used for automatically optimizing the rehabilitation training process of the elderly weak patient. According to the method, physical strength level, disease history and medicine use condition information of a patient are collected, physiological data of a rehabilitation person are monitored in real time in combination with various sensors, heart rate, blood oxygen, body temperature, myoelectric activity, fall detection, an accelerometer and a gyroscope, and a corresponding training scheme or adjustment suggestion is generated; the system outputs the physical health state of the rehabilitation person through the attention neural network; the body health state and the real-time data of the sensor serve as rule engine algorithm input, and then decision suggestions in rehabilitation training are made in real time; the method can reduce the health risk in training, and is especially suitable for personalized rehabilitation training of weak elderly patients.

Description

Technical Field

[0001] The invention discloses a safety monitoring method for rehabilitation training of frail elderly patients, belonging to the technical field of intelligent rehabilitation training systems. Background Art

[0002] With the increasing global aging phenomenon, rehabilitation training for frail elderly patients has become a major public health issue. Frailty refers to a decline in physical function, leading to a decrease in daily living ability and independence, and is particularly common among the elderly. Frail elderly patients often suffer from gait instability, muscle atrophy, and osteoporosis, which necessitate scientific monitoring and training programs during their rehabilitation. Traditional rehabilitation training for the elderly often relies on manual assessment and lacks real-time safety monitoring and immediate adjustment of training effects. Therefore, building a system that can monitor and adjust training intensity in real time is of great significance. Although some monitoring systems can currently collect real-time health data (such as heart rate, blood oxygen levels, and body temperature), they often lack personalized training programs and safety monitoring methods tailored to the specific needs of frail elderly patients. Furthermore, existing systems lack intelligence in data processing and real-time feedback, making it difficult to effectively combine multiple sensor data for health assessment and training intensity adjustments. Therefore, a safety monitoring method for rehabilitation training of frail elderly patients based on multiple sensor data and an attention neural network deep learning model is proposed. This method can comprehensively assess the patient's health status and accurately adjust the training plan based on real-time data, thereby improving the safety and effectiveness of rehabilitation training. Summary of the Invention

[0003] In order to overcome the above-mentioned deficiencies in the prior art, the present invention provides a safety monitoring method for rehabilitation training of frail elderly patients, which monitors the dangerous behaviors and status of rehabilitation personnel during rehabilitation training in real time. The technical solution of the present invention is as follows: A safety monitoring method for rehabilitation training of frail elderly patients, comprising the following steps: Step 1: Preliminary assessment of the patient's physical fitness level, medication use, and cardiovascular health status. Entropy weighting was used to determine the initial weights. Combined with expert ratings, a comprehensive score was calculated using the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation (FCE). This comprehensive score was then input into PCA for dimensionality reduction to three dimensions. K-means was then used to cluster the patients, resulting in three initial groupings: a "low-intensity aerobic group," a "resistance balance group," and a "functional group." Step 2: Real-time collection of various sensor data of rehabilitation patients; Step 3: Output the health status score of the rehabilitation personnel through the attention neural network; Step 4: Based on the preset rule engine algorithm and the mechanisms of condition matching, threshold determination, priority sorting and feedback control, auxiliary decision-making recommendations are implemented based on the health status score and the real-time data of the sensor.

[0004] Furthermore, the specific process of step 1 is as follows: 1.1. Initial Assessment and Training Intensity Setting Assessment content: The patient's physical level: including gait, balance, endurance, and muscle strength; Gait score: score of gait stability, stride length, etc., ranging from 0 to 10; Balance score: A score based on one-leg stand time or other balance ability tests, ranging from 0 to 10; Endurance score: through cardiopulmonary endurance test (such as maximum aerobic capacity, etc.), ranging from 0-10; Muscle strength score: assessed by grip strength, lower limb strength and other indicators, ranging from 0 to 10; Medical history: Understand whether the patient has common geriatric diseases such as diabetes, hypertension, and osteoporosis, and clarify whether there are chronic diseases or acute symptoms; score 0-10 based on whether the patient has diabetes, hypertension, osteoporosis and other common chronic diseases under control; Medication use: Consult a doctor to assess whether the patient is taking medication that may affect exercise ability; score 0-10 based on whether the patient is taking medication that affects exercise ability; Cardiovascular health status: Cardiovascular health is assessed through blood pressure, heart rate, and exercise electrocardiogram indicators; a comprehensive assessment of health indicators such as heart rate, blood pressure, and exercise electrocardiogram is performed with a score of 0-10.

[0005] 1.2. Multi-dimensional indicator weighting and clustering 1.2.1 Entropy Weight Method Initial Weight The patient assessment indicator matrix is ​​as follows: Where m is the number of patients, n is the number of evaluation indicators, X is the indicator matrix, is the original indicator matrix; Normalization formula: in is the specific normalized patient evaluation index matrix, m is the number of patients, n is the number of evaluation indicators, is a specific matrix value; Calculate the information entropy of the jth indicator: in is the exponential of the normalized evaluation matrix, is the specific normalized patient evaluation index matrix, m is the number of patients, n is the number of evaluation indicators, is the information entropy of the jth evaluation indicator; Entropy weight: in , for The overall sum, is the final entropy weight; 1.2.2 Modification of the Analytic Hierarchy Process (AHP) Constructing a judgment matrix composed of expert scores , calculate the eigenvector (maximum eigenvalue The corresponding normalized eigenvector) in : Judgment matrix given by experts, element represents the importance score of indicator j relative to indicator k, :The maximum characteristic root of the judgment matrix A, :and The corresponding normalized eigenvector is used as the weight obtained by AHP; Consistency check: in , : Random Index, which is a constant given in the table of the AHP method and corresponds to n; 1.2.3 Fuzzy Comprehensive Evaluation (FCE) Establish the fuzzy membership matrix of patients to evaluation indicators in : fuzzy membership matrix, represents the membership of the i-th patient on the j-th indicator, is the membership function; Comprehensive weight: in : Entropy weight method and AHP weight fusion coefficient, : Final comprehensive weight; Calculate the patient's composite score: in : comprehensive score of the i-th patient; 1.2.4 Comprehensive patient scoring and grouping Based on the results of entropy weight method, AHP method and fuzzy comprehensive evaluation, the comprehensive score of each patient is calculated. , and divide patients into appropriate training groups based on the scores; Grouping criteria: Low-intensity aerobic group ( ): Patients with good cardiovascular health and good physical fitness who are suitable for low-intensity aerobic training.

[0006] Resistance balance group ( ): Suitable for patients with mild muscle atrophy, poor balance or unstable walking, the training focuses on resistance and balance training.

[0007] Functional training group ( ): It is suitable for patients with poor physical strength and decreased basic activity ability, and the training focuses on restoring basic activity ability.

[0008] 1.3. Standardized training program Option 1: Low-intensity aerobic training Applicable population: 1) Mildly frail patients with good cardiovascular condition and no obvious bone and joint problems; 2) Patients without common geriatric diseases such as diabetes, hypertension, osteoporosis, and not taking medications that may affect exercise ability; 3) Patients with a certain level of physical fitness, able to perform continuous low-intensity exercise, and without symptoms of acute illness; Training content: Low-intensity walking training, light indoor cycling or aerobic training using an elliptical machine; Goal: Improve cardiopulmonary function, enhance basic endurance, and reduce the risk of cardiovascular disease; Training intensity: 15-30 minutes, 3-5 times a week, with the heart rate maintained between 50% and 60% of the maximum heart rate during training; Monitoring method: Regularly measure the patient's heart rate and blood oxygen level, and adjust the training intensity according to their response; Option 2: Combining resistance training with balance training Applicable population: 1) Patients with osteoporosis, muscle atrophy or unsteady walking; 2) Patients who are determined by a doctor to be able to perform resistance training and do not have serious cardiovascular problems; 3) Patients with a certain level of physical strength and are able to perform balance training and resistance training (including the use of light dumbbells and elastic bands); 4) Patients with no acute symptoms or contraindications in their medical history, such as severe hypertension or acute heart disease; Training content: Combine resistance training, light dumbbells, elastic bands and simple balance training, standing, squatting, and standing on one leg; Goal: Increase muscle strength, improve bone density, and enhance body balance and coordination; Training intensity: 2-3 times a week, about 30 minutes each time; resistance training uses a light load of 20%-30% of the maximum weight, 2-3 sets of each exercise, 8-12 repetitions per set; balance training focuses on practice time, 15 minutes each time; Monitoring method: Adjust the intensity by observing the patient's balance stability during training and the degree of muscle fatigue during resistance training; Option 3: Functional training Applicable population: 1) Patients with low physical strength, decreased basic activity ability, and difficulty standing or walking; 2) Patients with a history of chronic diseases (such as diabetes, hypertension, etc.), but these diseases have been controlled and do not affect functional training; 3) Patients can perform basic functional training such as standing, walking, and going up and down stairs under the guidance of a doctor, but attention should be paid to the patient's physical response and symptom changes; 4) Patients who are not taking medications that may affect exercise ability and do not have acute symptoms or serious cardiovascular problems; Training content: including functional movement training of standing up, walking, and going up and down stairs; Goal: To restore and improve daily living activities and increase the patient's independence; Training intensity: Depending on the patient's physical strength, each training session lasts 15-20 minutes, focusing on training basic motor skills; each functional training movement is repeated 10-15 times, gradually increasing the number and intensity; Monitoring method: Based on the patient's performance, physical response and recovery of activity during training, timely adjust the training content and intensity; Furthermore, the specific process of step 2 is: A variety of physiological data can be collected through sensors; 2.1. Heart Rate Sensor Sensor Type: Heart rate sensor embedded in smartwatch; Output data: Heart rate: the number of heartbeats per minute, reflecting the burden on the heart and the intensity of exercise; Heart rate variability: reflects the balance of the autonomic nervous system. A lower HRV indicates physical fatigue or excessive stress. Data significance: Heart rate: helps assess exercise intensity; during training, your heart rate should be kept within a certain percentage of your maximum heart rate (usually 50%-70%) to avoid overtraining; Heart rate variability: A high HRV generally indicates good heart health and recovery, while a low HRV can signal overtraining or a state of stress. 2.2. Blood Oxygen Sensor Sensor type: The blood oxygen sensor is an integrated device in the smartwatch; Output data: Blood oxygen saturation: indicates the concentration of oxygen in the blood, usually expressed as a percentage, with a normal range of 95%-100%); Data significance: Blood oxygen saturation: By monitoring blood oxygen levels, it can be determined whether the patient is in a state of hypoxia during exercise, especially for patients with poor cardiopulmonary function. Low blood oxygen may indicate that the training intensity is too high. 2.3. Body temperature sensor Sensor type: Body temperature sensor integrated into smartwatch; Output data: Body temperature: Real-time monitoring of the patient's body temperature changes to help detect possible overexertion or fever reactions during training; Data significance: Body temperature: A high body temperature may indicate that the training intensity is too high, causing the body to overheat, suggesting that the training plan needs to be adjusted; 2.4. Muscle electrical activity sensor Sensor type: Patch-type EMG sensors are attached to sportswear; Output data: Electromyographic signals: used to detect muscle activity and fatigue, reflecting muscle working intensity; Muscle activity level: real-time monitoring of the contraction intensity of specific muscle groups during training; Data significance: Electromyographic signals: used to assess the patient's muscle performance during resistance training, whether muscle fatigue has been reached, and to avoid overload training or muscle misuse; 2.5. Fall Detection Sensor Sensor Type: Built-in accelerometer and gyroscope in the belt; Output data: Fall events: Real-time detection and recording of whether a patient falls; Posture at the time of fall: Capture the angle and direction of the fall; Data significance: Fall detection: Falls are a common and serious problem for frail elderly patients. Through real-time monitoring, fall detection can issue timely alerts to reduce injuries. 2.6. Accelerometer and Gyroscope Sensor type: Three-axis accelerometer and gyroscope integrated into the smartwatch; Output data: Gait data: number of steps, cadence, and stride length, reflecting the patient's mobility and walking condition; Movement direction and angle: Monitor the patient's posture changes during balance training through a gyroscope; Acceleration data: can be used to analyze the patient's body movement patterns when walking, going up and down stairs, and squatting; Activity time: Real-time record of the total time the patient exercises; Rest time: records the time the patient spends at rest or in low-intensity activity; Data significance: Gait data: reflects the patient's motor ability and gait stability, and is often used to assess the walking ability and fall risk of frail patients; Movement direction and angle: helps monitor posture control during balance training, adjust training intensity in time, and avoid falls; Activity time: helps assess the total load of training and ensures the amount of exercise is moderate; Rest time: Avoid overtraining by monitoring rest time and ensure adequate recovery time; Furthermore, the specific process of step 3 is as follows: Deep learning technology is introduced to output an intuitive and rule-friendly health status variable based on sensor data such as heart rate, blood oxygen, gait, body temperature, electromyography, fall detection, and exercise time. The attention neural network is designed through a deep neural network, using a multi-layer perceptron and self-attention mechanism for data mapping. 3.1. Input sensor data preprocessing The input is a vector of sensor data: Gait data: stride length, cadence, gait symmetry Heart rate data: beats per minute, heart rate variability Blood oxygen data: SpO2 level Body temperature data: body temperature value EMG data: muscle activity signals Fall detection data: whether a fall occurred Exercise time and rest time: the duration of exercise and rest time The data vector collected from the sensor in real time at each moment is: in, : the length of each step, in meters; : The number of steps per minute, cadence, in steps / minute, steps / min; : Gait symmetry refers to the degree of balance between the left and right gaits, and is a value between 0 and 1, where 1 indicates complete symmetry and 0 indicates complete asymmetry; : heart rate, after standardization, unit is bpm; : Heart rate variability, after standardization, unit: ms; : Blood oxygen saturation, after standardization, unit: %; : body temperature, after standardization, unit: °C; : EMG signal amplitude, after normalization, unit: mV; : fall detection, binary state, 0 or 1; : acceleration, the normalized composite value of the three-axis acceleration, unit: g; : gyroscope angular velocity data, normalized attitude change rate, unit deg / s; is the standardized exercise duration and rest time, in minutes, min; Standardization method for each input quantity: in, Output data of the sensor over a period of time; The minimum value of the sensor output within a period of time; The maximum value of the sensor output within a period of time; is the normalized sensor data; 3.2. Attention Neural Network Definition Figure 3 This is the architecture diagram of the attention network, including the input layer, attention layer, feedforward network layer, and output layer; 3.2.1 The input layer is a sequence of sensor data: in, , : single time step data feature dimension, Data input for each time step; ; : original input data; t is the standard time step; 3.2.2 Core Mechanism of Attention Neural Network, Attention Mechanism): Compute query, key, value: in, , are weight matrices respectively; are the query, key, and value vectors obtained by transforming the weight matrix; : Input sensor data; Attention formula definition: in, is the dimension of the key vector; are the query, key, and value vectors obtained in 2.2; is an activation function that compresses any real vector into a probability distribution vector; Multi-head attention mechanism: Among them, the calculation of each head: : A separate weight matrix for each head; , h is the number of heads; To connect the overall expression of each head; is the overall weight expression of multi-head attention; : Feature dimension of a single time step data; trainable parameters for combining multi-head outputs; : is the linear transformation matrix; 3.2.3. Feedforward network definition: The feedforward network in the attention neural network consists of two fully connected layers; the output of each self-attention layer is further processed by two fully connected layers; the specific formula is as follows: in, is the weight matrix in the feedforward network, is the dimension of the hidden layer; and is the bias term; ReLU is the activation function; is the output of each self-attention layer; and is the linear transformation output; is the normalized output; is the feedforward network function; LayerNorm() is the layer normalization function; 3.2.4 Attention Neural Network Output, Single Layer): After multiple layers of self-attention and feedforward network processing, the final output representation It is passed to a fully connected layer for the final health status prediction; the output value is compressed to the range of [0,1] through the Sigmoid activation function: in, is the weight matrix of the output layer, is the bias term; Is a standard activation function; Final Output represents the health status score of the rehabilitation personnel at time t, ranging from 0 to 1, where values ​​close to 1 indicate a good health status and values ​​close to 0 indicate a poor health status; Furthermore, the specific process of step 4 is as follows: Real-time monitoring feedback typically takes precedence over adjustments to rehabilitation training plans, as ensuring the patient's physical condition and safety is the primary task of the rehabilitation training system. The rule engine algorithm uses the health status score output by the attention neural network and real-time data from heart rate sensors, blood oxygen sensors, body temperature sensors, muscle electrical activity sensors, fall detection sensors, accelerometers, and gyroscopes as decision-making inputs. Based on the rule engine and these inputs, the system provides real-time feedback and takes necessary preventive measures. According to different activity scenarios, the rule engine algorithm is used to make decisions. The specific priorities and decision rules are as follows: Figure 2 As shown: 4.1. Low Heart Rate Warning (Low-Intensity Aerobic Group, Resistance Balance Group, Functional Group) Real-time feedback conditions: when the health index (H) is less than 0.2 and the blood oxygen saturation is less than 85%; Decision Recommendation: Stop training and call for emergency response from medical personnel; 4.2. High Heart Rate Warning (Low-Intensity Aerobic Group, Resistance Balance Group, Functional Group) Real-time feedback conditions: when the health index is between 0.2 and 0.4, and the heart rate exceeds 70% of the maximum safe range; Decision-making recommendations: It is recommended to reduce training intensity and rest; 4.3. Muscle Fatigue Warning (Low-Intensity Aerobic Group, Balanced Resistance Group) Real-time feedback conditions: When the health index is between 0.4 and 0.6, and the electromyographic signal shows that the muscle activity level is too high; Decision-making recommendations: It is recommended to reduce the intensity and number of resistance training; 4.4. Body temperature rise warning (low-intensity aerobic group, resistance balance group) Real-time feedback conditions: when the health index is between 0.6 and 0.8 and the body temperature is too high; Decision-making recommendations: It is recommended to pause training, replenish water and adjust the ambient temperature; 4.5. Gait stability test (functional group) Real-time feedback conditions: When the health index is greater than 0.8 and there is a large deviation in cadence; Decision suggestion: It is recommended to continue the current training program, but pay attention to gait stability; 4.6. Fall Event Warning (Low-Intensity Aerobic Group, Resistance Balance Group, Functional Group) Real-time feedback conditions: when the fall detection sensor triggers a fall event; Decision suggestion: Immediately sound an alarm, record the posture and direction of the fall, and activate the emergency contact mechanism; 4.7. Low Heart Rate Variability Warning (Low-Intensity Aerobic Group, Resistance Balance Group, Functional Group) Real-time feedback conditions: when heart rate variability (HRV) is lower than 0.3; Decision-making recommendations: It is recommended to increase recovery time and perform deep breathing relaxation training; 4.8. Hypoxemia Warning (Low-Intensity Aerobic Group, Resistance Balance Group, Functional Group) Real-time feedback conditions: when blood oxygen saturation is lower than 90% and health index is less than 0.5; Decision-making recommendations: It is recommended to reduce the intensity of aerobic exercise and perform intermittent breathing training; 4.9. Overtraining Warning (Low-Intensity Aerobic Group, Balanced Resistance Group) Real-time feedback conditions: when the activity time exceeds 70% and the rest time is less than 20%; Decision-making recommendations: It is recommended to increase rest time to avoid overtraining; 4.10. Insufficient Resistance Training Load Warning (Balanced Resistance Group) Real-time feedback conditions: When the electromyographic signal shows that the muscle activity level is too low and the health index is greater than 0.5; Decision-making suggestion: It is recommended to appropriately increase the resistance training load to 25% of the maximum load; 4.11. High Heart Rate Warning (Low-Intensity Aerobic Group) Real-time feedback conditions: When the heart rate continuously exceeds 85% of the maximum heart rate and the health index is less than 0.5; Decision suggestion: It is recommended to immediately reduce the training intensity and rest for 5 minutes; 4.12. Hyperthermia Warning (Low-Intensity Aerobic Group, Resistance Balance Group, Functional Group) Real-time feedback conditions: when body temperature is too high and blood oxygen saturation is lower than 95%; Decision suggestion: It is recommended to stop training and cool down to avoid overheating; 4.13. Muscle fatigue warning, training reduction (low-intensity aerobic group, resistance balance group) Real-time feedback conditions: When the electromyographic signal continuously exceeds the fatigue threshold and the health index is lower than 0.3; Decision-making recommendations: It is recommended to reduce the number of training sessions and monitor muscle fatigue; 4.14. Balance Training Recommendations (Resistance Balance Group) Real-time feedback conditions: when the health index is greater than 0.5 and no fall event occurs; Decision-making suggestion: It is recommended to try moderate-intensity balance training and single-leg standing training; 4.15. Training Intensity Adjustment Recommendations (Low-Intensity Aerobic Group, Resistance Balance Group, Functional Group) Real-time feedback conditions: when heart rate variability is high and health index is greater than 0.7; Decision-making suggestion: It is recommended to appropriately increase the difficulty of training, increase resistance and intensity; 4.16. Gait Instability Warning (Resistance Balance Group, Functional Group) Real-time feedback condition: when the gait frequency variance exceeds 0.3; Decision-making suggestion: It is recommended to conduct special balance training and stand on one leg to enhance posture control ability; 4.17. Blood oxygen decreasing trend warning (functional group) Real-time feedback conditions: When blood oxygen saturation continues to decrease and its change slope is less than -0.01; Decision recommendations: It is recommended to check the patient's breathing pattern and continuously monitor lung function; Beneficial Effects: 1. Comprehensive Health Assessment: This method collects multiple physiological data in real time, such as heart rate, blood oxygen levels, body temperature, electromyographic activity, and fall detection, and applies an attention-based neural network model for data processing and feature extraction, generating an intuitive health score. This score accurately reflects the patient's physical condition and provides a scientific basis for developing personalized rehabilitation plans. 2. Intelligent Training Intensity Adjustment: This method leverages the health score output by the attention-based neural network model to adjust the intensity and approach of rehabilitation training in real time through a rules engine. When the system detects a health issue (such as an elevated heart rate, decreased blood oxygen levels, or elevated body temperature), it automatically triggers an alarm and recommends appropriate adjustments to training intensity based on pre-defined rules to ensure patient safety. 3. Real-Time Monitoring and Feedback: This method continuously tracks the patient's movement and physiological data and, through an immediate feedback mechanism, provides necessary adjustments to the patient or medical staff, such as increasing rest time or adjusting training intensity. This mechanism helps prevent overtraining or inappropriate training, reducing the risk of sports injuries. 4. Comprehensive Rehabilitation Training Program Design: Based on an assessment of the patient's physical condition, medical history, and medication use, this method can tailor rehabilitation training programs for different types of frail elderly patients. Whether it's low-intensity aerobic exercise, a combination of resistance and balance training, or functional training, this method can be flexibly adjusted to help patients achieve optimal rehabilitation outcomes. 5. Multi-source data fusion: By integrating multiple sensors such as heart rate, blood oxygen, body temperature, electromyographic signals, fall detection, accelerometers, and gyroscopes, this method can comprehensively monitor various physiological indicators of patients during the rehabilitation process. Combined with deep learning algorithms for comprehensive analysis, this method can obtain more accurate health status assessments, enhancing the system's intelligence and real-time responsiveness. 6. Fall warning and emergency response mechanism: This method has a fall detection function that can detect whether a patient has fallen in real time and immediately initiate emergency response procedures. The system can also analyze the angle and direction of the fall and record the specific circumstances of the fall, providing data support for subsequent health analysis and preventive measures. 7. Personalized training adjustment and adaptive management: By utilizing attention neural network technology, this method can dynamically adjust rehabilitation training plans based on the patient's real-time data, ensuring that the training intensity matches the patient's health status. For patients with poor physical strength or poor health, the system can intelligently adjust the training load to avoid the risk of overtraining while maintaining training results. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 The present invention provides an overall technical flow chart of a safety monitoring method for rehabilitation training of frail elderly patients; Figure 2 A diagram illustrating the priority of preset rules of the decision engine algorithm in step 4; Figure 3This is a conceptual diagram of the Transformer algorithm principle; Figure 4 This is a diagram showing the actual application of the Transformer algorithm. DETAILED DESCRIPTION

[0010] Figure 1 As shown, a safety monitoring method for rehabilitation training of frail elderly patients includes the following steps: Step 1: Preliminary assessment of the patient's physical fitness level, medication use, and cardiovascular health status. Entropy weighting was used to determine the initial weights. Combined with expert ratings, a comprehensive score was calculated using the analytic hierarchy process (AHP) and fuzzy comprehensive evaluation (FCE). This comprehensive score was then input into PCA for dimensionality reduction to three dimensions. K-means was then used to cluster the patients, resulting in three initial groupings: a "low-intensity aerobic group," a "resistance balance group," and a "functional group." Step 2: Real-time collection of various sensor data of rehabilitation patients; Step 3: Output the health status score of the rehabilitation personnel through the attention neural network; Step 4: Implement auxiliary decision-making recommendations based on the health status score and real-time sensor data based on the preset rule engine algorithm and mechanisms such as condition matching, threshold determination, priority sorting, and feedback control; 2. The specific process of step 1 is: 1. Initial assessment and training intensity setting Assessment content: When a patient is admitted to the hospital, an initial assessment is performed, including the following: Physical fitness level: Gait testing showed that the patient's stride length was 40 cm, the cadence was 0.8 steps / second, and the gait symmetry was 0.95; Gait score: 8 (stable gait, good stride) Balance score: 7 (standing on one leg for a long time, good balance) Endurance score: 6 (good aerobic endurance, but not top-tier) Muscle strength score: 7 (normal grip strength, good lower limb strength) Medical history score: 8 (no major diseases such as diabetes, hypertension, etc., in good health) Drug use: The drug use score of the patient was 9 (no drug interference, no drug affecting exercise ability) Cardiovascular health status: Blood pressure, heart rate, and exercise electrocardiogram were measured, and the patient's resting heart rate was 75 beats per minute. Exercise electrocardiogram showed good heart health with no significant abnormalities. Cardiovascular health score: 8 (heart rate, blood pressure, and electrocardiogram all showed normal) 2. Multi-dimensional indicator weighting and clustering 2.1 Initial weight of entropy weight method The patient assessment indicator matrix is ​​as follows: The sum of the data in each column (including 5 historical patients and new patients, a total of m=6): Normalization: Calculate the information entropy of the jth indicator: 2.2 Modification of the Analytic Hierarchy Process (AHP) Constructing a judgment matrix composed of expert scores , calculate the eigenvector (maximum eigenvalue The corresponding normalized eigenvector) Fusion coefficient =0.8, final weight Enter the value: 2.3 Fuzzy Comprehensive Evaluation (FCE) Establish the fuzzy membership matrix of patients to evaluation indicators Comprehensive weight: ,The patients were suitable for the low-intensity aerobic group; Low-intensity aerobic training Applicable population: mildly frail patients with good cardiovascular condition and no obvious bone and joint problems; Training content: Low-intensity walking training, light indoor cycling or aerobic training using an elliptical machine; Goal: Improve cardiopulmonary function, enhance basic endurance, and reduce the risk of cardiovascular disease; Training intensity: 15-30 minutes, 3-5 times a week, with the heart rate maintained between 50% and 60% of the maximum heart rate during training; Monitoring method: Regularly measure the patient's heart rate and blood oxygen level, and adjust the training intensity according to their response; 3. The specific process of step 2 is: Smartwatches and sensors collect patients’ physiological data in real time, including: 1. Heart rate sensor Sensor Type: Heart rate sensor embedded in smartwatch; Output data: The patient's heart rate was monitored in real time. During the training process, the patient's heart rate was 80 beats per minute, close to 58% of the maximum heart rate, indicating that the patient's training intensity was within a safe range. 2. Blood oxygen sensor, pulse oximeter Sensor type: The blood oxygen sensor is an integrated device in the smartwatch; Output data: Blood oxygen saturation, SpO2, was 92%, which is within the normal range, indicating that the patient's cardiopulmonary fitness is good; 3. Body temperature sensor Sensor type: Body temperature sensor integrated into smartwatch; Output data: The patient's body temperature was monitored in real time; it was 36.8°C, close to the normal temperature range, indicating that the training did not cause excessive load; 4. Muscle electrical activity sensor Sensor type: Patch-type EMG sensors are attached to sportswear; Output data: Real-time monitoring of the patient's muscle activity; the electromyographic signal = 0.15 mV, with small fluctuations, indicates that the muscles have not reached a fatigue state and the training intensity is appropriate; 5. Fall detection sensor Sensor Type: Built-in accelerometer and gyroscope in the belt; Output data: Fall events: No falls occurred, and the sensor showed that the patient remained well stable during the training; 6. Accelerometer and gyroscope Sensor type: Three-axis accelerometer and gyroscope integrated into the smartwatch; Output data: Gait data: Gait data showed a cadence of 80 steps / min, a stride of 60 cm, and a gait symmetry of 0.9, indicating that the patient had good walking ability; Gyroscope data: angular velocity = 2.0 deg / s; Acceleration data: acceleration = 0.5 g; Exercise time: The patient's total exercise time was recorded through monitoring with accelerometers and motion sensors; the patient's rehabilitation training time was 30 minutes; Rest time: The patient's rest or low-intensity activity time during rehabilitation training is 15 minutes. The system will record this time and remind the rehabilitation therapist to adjust the training intensity or rest time; 4. The safety monitoring method for rehabilitation training of frail elderly patients according to claim 1, wherein the specific process of step 3 is as follows: Deep learning technology is introduced to output an intuitive and rule-friendly health status variable based on sensor data such as heart rate, blood oxygen, gait, body temperature, electromyography, fall detection, and exercise time. The attention neural network is designed through a deep neural network, using multiple encoders and decoders and a self-attention mechanism for data mapping. 1. Input sensor data preprocessing The data vector collected from the sensor in real time at each moment is: Among them, these are the sensor data of patient A at time t. After normalization, the input feature vector is obtained, with a total of 13 features and a dimension of 13); 2. Definition of Attention Neural Network 2.1 The input layer is a sequence of sensor data: in, , : Feature dimension of single time step data; ; : original input data; t is the standard time step; 2.2 Multi-head Attention Mechanism: Among them, the calculation of each head: : A separate weight matrix for each head; , the number of attention heads is h=4; is 13, that is, the input data dimension of each time step is 13; Calculation query (Query), key (Key), value (Value): in, , are weight matrices respectively; are the query, key, and value vectors obtained by transforming the weight matrix; : Input sensor data; By matrix multiplication we get: Attention score calculation: Similarly, for other heads, we calculate the corresponding attention output: Merge the multi-head outputs and map them to the final dimension through a fully connected layer, ); 2.3. Feedforward network definition: The feedforward network in the attention neural network consists of two fully connected layers; the output of each self-attention layer is further processed by two fully connected layers; the specific formula is as follows: in, is the weight matrix in the feedforward network, is the dimension of the hidden layer; and is the bias term; ReLU is the activation function; is the output of each self-attention layer; and is the linear transformation output; is the normalized output; is the feedforward network function; LayerNorm() is the layer normalization function; The attention weight is obtained by normalization: 2.4 Attention Neural Network Output, Single Layer: After multiple layers of self-attention and feedforward network processing, the final output representation It is passed to a fully connected layer for the final health status prediction; the output value is compressed to the range of [0,1] through the Sigmoid activation function: in, is the weight matrix of the output layer, is the bias term; Is a standard activation function; Final Output represents the health status score of the rehabilitation personnel at time t, ranging from 0 to 1, where values ​​close to 1 indicate a good health status and values ​​close to 0 indicate a poor health status; Figure 4 The curve graph during algorithm training shows that the effectiveness gradually reaches more than 90% as the number of iterations increases.

[0011] 5. The specific process of step 4 is: Real-time monitoring feedback typically takes precedence over adjustments to rehabilitation training plans, as ensuring the patient's physical condition and safety is the primary task of the rehabilitation training system. The rule engine algorithm uses the health status score of the attention neural output and real-time data from heart rate sensors, blood oxygen sensors, body temperature sensors, muscle electrical activity sensors, fall detection sensors, accelerometers, and gyroscopes as decision-making inputs. Based on the rule engine and these inputs, the system provides real-time feedback and takes necessary preventive measures. Based on different activity scenarios, rule engine algorithms are used to make decisions. Specific priorities and decision rules are as follows: Rule 1: Low Heart Rate Warning Condition: When health index and blood oxygen saturation is less than 85%; Real-time data: Patient A's health status score is , and the blood oxygen saturation is 92%; Judgment: This condition is not satisfied because and blood oxygen saturation is higher than 85%; Decision: No action is needed, continue to observe. Rule 2: High Heart Rate Warning Conditions: When the health index is between 0.2 and 0.4, and the heart rate exceeds 70% of the maximum safe range; Real-time data: Patient A's health status score is , and heart rate is 80 bpm; Judgment: This condition is not satisfied because More than 0.4, and the heart rate is 80 bpm, which is below the maximum safe range of 84 bpm; Decision: No action is needed, continue to observe. Rule 3: Muscle Fatigue Warning Conditions: When the health index is between 0.4 and 0.6, and the electromyographic signal shows that the muscle activity level is too high; Real-time data: Patient A's health status score is , and the EMG signal amplitude is 0.15 mV; Judgment: This condition is not satisfied because It exceeds 0.6, and the EMG signal amplitude is low; Decision: No action is needed, continue to observe. Rule 4: Warning of rapid rise in body temperature Conditions: When the health index is between 0.6 and 0.8 and the body temperature is too high; Real-time data: Patient A's health status score is , and a body temperature of 36.8°C; Judgment: This condition is not met because the body temperature is within the normal range, 36.8°C), and there is no sign of hyperthermia; Decision: No action is needed, continue to observe. Rule 5: Gait Stability Check Condition: When the health index is greater than 0.8 and the cadence has a large deviation; Real-time data: Patient A's health status score is , the cadence was 80 steps / min, with no significant deviation; Judgment: This condition is not satisfied because Less than 0.8, and no significant deviation in cadence; Decision: Continue to observe, no need to adjust the training plan; Rule 6: Fall Incident Warning Condition: When the fall detection sensor triggers a fall event; Real-time data: Patient A did not fall, and the fall event was 0); Judgment: This condition is not met because the patient did not fall; Decision: No action is needed, continue to observe. Rule 7: Overtraining Warning Condition: When the activity time exceeds 70% and the rest time is less than 20%; Real-time data: Patient A's activity time is 30 minutes, and rest time is 15 minutes, with standardized values ​​of 0.6 and 0.3); Calculation: activity time ratio is 67%, rest time ratio is 33%; Judgment: This condition is not met because the activity time is 67%, which is lower than 70%; Decision: No action is needed, continue with the current training plan. Rule 8: Training Intensity Adjustment Recommendations Conditions: When heart rate variability is high and health index ; Real-time data: Patient A's heart rate variability is 45 ms, normalized to 0.55, and his health status score is 0.692; Judgment: This condition is not satisfied because Less than 0.7; Decision: No need to increase training intensity, continue to observe.

[0012] Finally, it should be noted that the above-mentioned c is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein. Any modifications, replacements, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A safety monitoring method for rehabilitation training of frail elderly patients, characterized in that: The following steps are involved: Step 1: Preliminary assessment of the patient's physical fitness level, medication use, and cardiovascular health status. Entropy weighting was used to initially determine weights. Combined with expert ratings, a comprehensive score was calculated using the Analytic Hierarchy Process (AHP) and Fuzzy Comprehensive Evaluation (FCE). This comprehensive score was then input into PCA for dimensionality reduction to three dimensions. K-means was then used to cluster the patients, resulting in three initial groupings: a "low-intensity aerobic group," a "resistance balance group," and a "functional group." Step 2: Real-time collection of various sensor data of rehabilitation patients; Step 3: Output the health status score of the rehabilitation personnel through the attention neural network; Step 4: Based on the preset rule engine algorithm and the mechanisms of condition matching, threshold determination, priority sorting and feedback control, auxiliary decision-making recommendations are implemented based on the health status score and the real-time data of the sensor.

2. A safety monitoring method for rehabilitation training of frail elderly patients according to claim 1, characterized in that: The specific process of step 1 is: Step 1.1 Preliminary assessment and training intensity setting, Assessment content: The patient's physical level: including gait, balance, endurance, and muscle strength; Gait score: score of gait stability, stride length, etc., ranging from 0 to 10; Balance score: A score based on one-leg stand time or other balance ability tests, ranging from 0 to 10; Endurance score: through cardiopulmonary endurance test (such as maximum aerobic capacity, etc.), ranging from 0-10; Muscle strength score: assessed by grip strength, lower limb strength and other indicators, ranging from 0 to 10; Medical history: Understand whether the patient has common geriatric diseases such as diabetes, hypertension, and osteoporosis, and clarify whether there are chronic diseases or acute symptoms; score 0-10 based on whether the patient has diabetes, hypertension, osteoporosis and other common chronic diseases under control; Medication use: Consult a doctor to assess whether the patient is taking medication that may affect exercise ability; score 0-10 based on whether the patient is taking medication that affects exercise ability; Cardiovascular health: Cardiovascular health is assessed through blood pressure, heart rate, and exercise electrocardiogram indicators; a comprehensive assessment of health indicators such as heart rate, blood pressure, and exercise electrocardiogram is performed with a score of 0-10; Step 1.2 Multi-dimensional indicator weighting and clustering, Step 1.2.1 Entropy weight method initial weight, The patient assessment indicator matrix is ​​as follows: Where m is the number of patients, n is the number of evaluation indicators, X is the indicator matrix, is the original indicator matrix; Normalization formula: in is the specific normalized patient evaluation index matrix, m is the number of patients, n is the number of evaluation indicators, is a specific matrix value; Calculate the information entropy of the jth indicator: in is the exponential of the normalized evaluation matrix, is the specific normalized patient evaluation index matrix, m is the number of patients, n is the number of evaluation indicators, is the information entropy of the jth evaluation indicator; Entropy weight: in , for The overall sum, is the final entropy weight; Step 1.2.2: Modification of the Analytic Hierarchy Process (AHP). Constructing a judgment matrix composed of expert scores , calculate the eigenvector (maximum eigenvalue Corresponding normalized eigenvectors): ； in : Judgment matrix given by experts, element represents the importance score of indicator j relative to indicator k, :The maximum characteristic root of the judgment matrix A, :and The corresponding normalized eigenvector is used as the weight obtained by AHP; Consistency check: ; in , : Random Index, which is a constant given in the table of the AHP method and corresponds to n; Step 1.2.3 Fuzzy Comprehensive Evaluation (FCE), Establish the fuzzy membership matrix of patients to evaluation indicators: ; in : fuzzy membership matrix, represents the membership of the i-th patient on the j-th indicator, is the membership function; Comprehensive weight: ; in : Entropy weight method and AHP weight fusion coefficient, : Final comprehensive weight; Calculate the patient's composite score: ; in : comprehensive score of the i-th patient; Step 1.2.4 Comprehensive scoring and grouping of patients, Based on the results of entropy weight method, AHP method and fuzzy comprehensive evaluation, the comprehensive score of each patient is calculated. , and divide patients into appropriate training groups based on the scores; Grouping criteria: Low-intensity aerobic group ( ): Patients with good cardiovascular health and good physical fitness who are suitable for low-intensity aerobic training, Resistance balance group ( ): Suitable for patients with mild muscle atrophy, poor balance or unstable walking. The training focuses on resistance and balance training. Functional training group ( ): Suitable for patients with poor physical strength and decreased basic activity ability. The training focuses on restoring basic activity ability. Step 1.3 Standardize the training program, Option 1: Low-intensity aerobic training. Applicable population: 1) Mildly frail patients with good cardiovascular condition and no obvious bone and joint problems; 2) Patients without common geriatric diseases such as diabetes, hypertension, osteoporosis, and not taking medications that may affect exercise ability; 3) Patients with a certain level of physical fitness, able to perform continuous low-intensity exercise, and without symptoms of acute illness; Training content: Low-intensity walking training, light indoor cycling or aerobic training using an elliptical machine; Goal: Improve cardiopulmonary function, enhance basic endurance, and reduce the risk of cardiovascular disease; Training intensity: 15-30 minutes, 3-5 times a week, with the heart rate maintained between 50% and 60% of the maximum heart rate during training; Monitoring method: Regularly measure the patient's heart rate and blood oxygen level, and adjust the training intensity according to their response; Option 2: Combination of resistance training and balance training. Applicable population: 1) Patients with osteoporosis, muscle atrophy or unsteady walking; 2) Patients who are determined by a doctor to be able to perform resistance training and do not have serious cardiovascular problems; 3) Patients with a certain level of physical strength and are able to perform balance training and resistance training (including the use of light dumbbells and elastic bands); 4) Patients with no acute symptoms or contraindications in their medical history, such as severe hypertension or acute heart disease; Training content: Combine resistance training, light dumbbells, elastic bands and simple balance training, standing, squatting, and standing on one leg; Goal: Increase muscle strength, improve bone density, and enhance body balance and coordination; Training intensity: 2-3 times a week, about 30 minutes each time; resistance training uses a light load of 20%-30% of the maximum weight, 2-3 sets of each exercise, 8-12 repetitions per set; balance training focuses on practice time, 15 minutes each time; Monitoring method: Adjust the intensity by observing the patient's balance stability during training and the degree of muscle fatigue during resistance training; Option 3: Functional training, Applicable population: 1) Patients with low physical strength, decreased basic activity ability, and difficulty standing or walking; 2) Patients with a history of chronic diseases (such as diabetes, hypertension, etc.), but these diseases have been controlled and do not affect functional training; 3) Patients can perform basic functional training such as standing, walking, and climbing stairs under the guidance of a doctor, but attention should be paid to the patient's physical response and symptom changes; 4) Patients who are not taking medications that may affect exercise ability and do not have acute symptoms or serious cardiovascular problems; Training content: including functional movement training of standing up, walking, and going up and down stairs; Goal: To restore and improve daily living activities and increase the patient's independence; Training intensity: Depending on the patient's physical strength, each training session lasts 15-20 minutes, focusing on training basic motor skills; each functional training movement is repeated 10-15 times, gradually increasing the number and intensity; Monitoring method: Adjust the training content and intensity in a timely manner based on the patient's performance, physical response and recovery of activity ability during training.

3. A safety monitoring method for rehabilitation training of frail elderly patients according to claim 1, characterized in that: The specific process of step 2 is: A variety of physiological data can be collected through sensors; Step 2.1 Heart rate sensor, Sensor Type: Heart rate sensor embedded in smartwatch; Output data: Heart rate: the number of heartbeats per minute, reflecting the burden on the heart and the intensity of exercise; Heart rate variability: reflects the balance of the autonomic nervous system. A lower HRV indicates physical fatigue or excessive stress. Data significance: Heart rate: helps assess exercise intensity. During training, your heart rate should be kept within a certain percentage of your maximum heart rate, usually 50%-70%, to avoid overtraining. Heart rate variability: A high HRV generally indicates good heart health and recovery, while a low HRV can signal overtraining or a state of stress. Step 2.2 Blood oxygen sensor, Sensor type: The blood oxygen sensor is an integrated device in the smartwatch; Output data: Blood oxygen saturation: indicates the concentration of oxygen in the blood, usually expressed as a percentage, with a normal range of 95%-100%; Data significance: Blood oxygen saturation: By monitoring blood oxygen levels, it can be determined whether the patient is in a state of hypoxia during exercise, especially for patients with poor cardiopulmonary function. Low blood oxygen may indicate that the training intensity is too high. Step 2.3 Body temperature sensor, Sensor type: Body temperature sensor integrated into smartwatch; Output data: Body temperature: Real-time monitoring of the patient's body temperature changes to help detect possible overexertion or fever reactions during training; Data significance: Body temperature: A high body temperature may indicate that the training intensity is too high, causing the body to overheat, suggesting that the training plan needs to be adjusted; Step 2.4 Muscle electrical activity sensor, Sensor type: Patch-type EMG sensors are attached to sportswear; Output data: Electromyographic signals: used to detect muscle activity and fatigue, reflecting muscle working intensity; Muscle activity level: real-time monitoring of the contraction intensity of specific muscle groups during training; Data significance: Electromyographic signals: used to assess the patient's muscle performance during resistance training, whether muscle fatigue has been reached, and to avoid overload training or muscle misuse; Step 2.5 Fall Detection Sensor Sensor Type: Built-in accelerometer and gyroscope in the belt; Output data: Fall events: Real-time detection and recording of whether a patient falls; Posture at the time of fall: Capture the angle and direction of the fall; Data significance: Fall detection: Falls are a common and serious problem for frail elderly patients. Through real-time monitoring, fall detection can issue timely alerts to reduce injuries. Step 2.6 Accelerometer and Gyroscope Sensor type: Three-axis accelerometer and gyroscope integrated into the smartwatch; Output data: Gait data: number of steps, cadence, and stride length, reflecting the patient's mobility and walking condition; Movement direction and angle: Monitor the patient's posture changes during balance training through a gyroscope; Acceleration data: can be used to analyze the patient's body movement patterns when walking, going up and down stairs, and squatting; Activity time: Real-time record of the total time the patient exercises; Rest time: records the time the patient spends at rest or in low-intensity activity; Data significance: Gait data: reflects the patient's motor ability and gait stability, and is often used to assess the walking ability and fall risk of frail patients; Movement direction and angle: helps monitor posture control during balance training, adjust training intensity in time, and avoid falls; Activity time: helps assess the total load of training and ensures the amount of exercise is moderate; Rest periods: Avoid overtraining by monitoring your rest periods and ensure adequate recovery time.

4. A safety monitoring method for rehabilitation training of frail elderly patients according to claim 1, characterized in that: The specific process of step 3 is as follows: Deep learning technology is introduced to output an intuitive and rule-friendly health status variable based on sensor data such as heart rate, blood oxygen, gait, body temperature, electromyography, fall detection, and exercise time. The attention neural network is designed through a deep neural network, using a multi-layer perceptron and self-attention mechanism for data mapping. Step 3.1 Input sensor data preprocessing, The input is a vector of sensor data: Gait data: stride length, cadence, gait symmetry Heart rate data: beats per minute, heart rate variability Blood oxygen data: SpO2 level Body temperature data: body temperature value EMG data: muscle activity signals Fall detection data: whether a fall occurred Exercise time and rest time: the duration of exercise and rest time The data vector collected from the sensor in real time at each moment is: in, : the length of each step, in meters; : The number of steps per minute, cadence, in steps / minute, steps / min; : Gait symmetry refers to the degree of balance between the left and right gaits, and is a value between 0 and 1, where 1 indicates complete symmetry and 0 indicates complete asymmetry; : heart rate, after standardization, unit is bpm; : Heart rate variability, after standardization, unit: ms; : Blood oxygen saturation, after standardization, unit: %; : body temperature, after standardization, unit: °C; : EMG signal amplitude, after normalization, unit: mV; : fall detection, binary state, 0 or 1; : acceleration, the normalized composite value of the three-axis acceleration, unit: g; : gyroscope angular velocity data, normalized attitude change rate, unit deg / s; is the standardized exercise duration and rest time, in minutes, min; Standardization method for each input quantity: in, Output data of the sensor over a period of time; The minimum value of the sensor output within a period of time; The maximum value of the sensor output within a period of time; is the normalized sensor data; Step 3.2 Attention neural network definition, Figure 3 shows the architecture of the attention network, which includes the input layer, attention layer, feedforward network layer, and output layer; Step 3.2.1 The input layer is a sequence of sensor data: in, , : single time step data feature dimension, Data input for each time step; ; : original input data; t is the standard time step; Step 3.2.2 The core mechanism of attention neural network, attention mechanism, Compute query, key, value: in, , are weight matrices respectively; are the query, key, and value vectors obtained by transforming the weight matrix; : Input sensor data; Attention formula definition: in, is the dimension of the key vector; are the query, key, and value vectors obtained in 2.2; is an activation function that compresses any real vector into a probability distribution vector; Multi-head attention mechanism: Among them, the calculation of each head: : A separate weight matrix for each head; , h is the number of heads; To connect the overall expression of each head; is the overall weight expression of multi-head attention; : Feature dimension of a single time step data; trainable parameters for combining multi-head outputs; : is the linear transformation matrix; Step 3.2.3 Feedforward network definition: The feedforward network in the attention neural network consists of two fully connected layers; the output of each self-attention layer is further processed by two fully connected layers; the specific formula is as follows: in, is the weight matrix in the feedforward network, is the dimension of the hidden layer; and is the bias term; ReLU is the activation function; is the output of each self-attention layer; and is the linear transformation output; is the normalized output; is the feedforward network function; LayerNorm() is the layer normalization function; Step 3.2.4 Attention neural network output: After multiple layers of self-attention and feedforward network processing, the final output representation It is passed to a fully connected layer for the final health status prediction; the output value is compressed to the range of [0,1] through the Sigmoid activation function: in, is the weight matrix of the output layer, is the bias term; Is a standard activation function; Final Output It represents the health status score of the rehabilitation personnel at time t, ranging from 0 to 1, where values ​​close to 1 indicate better health status and values ​​close to 0 indicate worse health status.

5. A safety monitoring method for rehabilitation training of frail elderly patients according to claim 1, characterized in that: The specific process of step 4 is as follows: Real-time monitoring feedback typically takes precedence over adjustments to rehabilitation training plans, as ensuring the patient's physical condition and safety is the primary task of the rehabilitation training system. The rule engine algorithm uses the health status score output by the attention neural network and real-time data from heart rate sensors, blood oxygen sensors, body temperature sensors, muscle electrical activity sensors, fall detection sensors, accelerometers, and gyroscopes as decision-making inputs. Based on the rule engine and these inputs, the system provides real-time feedback and takes necessary preventive measures. Based on different activity scenarios, rule engine algorithms are used to make decisions. Specific priorities and decision rules are as follows: Step 4.1 Low heart rate warning (low intensity aerobic group, resistance balance group, functional group), Real-time feedback conditions: when the health index is less than 0.2 and the blood oxygen saturation is less than 85%; Decision Recommendation: Stop training and call for emergency response from medical personnel; Step 4.2 High heart rate warning (low-intensity aerobic group, resistance balance group, functional group), Real-time feedback conditions: when the health index is between 0.2 and 0.4, and the heart rate exceeds 70% of the maximum safe range; Decision-making recommendations: It is recommended to reduce training intensity and rest; Step 4.3 Muscle fatigue warning (low-intensity aerobic group, resistance balance group), Real-time feedback conditions: When the health index is between 0.4 and 0.6, and the electromyographic signal shows that the muscle activity level is too high; Decision-making recommendations: It is recommended to reduce the intensity and number of resistance training; Step 4.4: Warning of rapid rise in body temperature (low-intensity aerobic group, resistance balance group). Real-time feedback conditions: when the health index is between 0.6 and 0.8 and the body temperature is too high; Decision-making recommendations: It is recommended to pause training, replenish water and adjust the ambient temperature; Step 4.5 Gait stability test (functional group), Real-time feedback conditions: When the health index is greater than 0.8 and the cadence has a large deviation; Decision suggestion: It is recommended to continue the current training program, but pay attention to gait stability; Step 4.6 Fall event warning (low-intensity aerobic group, resistance balance group, functional group), Real-time feedback conditions: when the fall detection sensor triggers a fall event; Decision suggestion: Immediately sound an alarm, record the posture and direction of the fall, and activate the emergency contact mechanism; Step 4.7: Low heart rate variability warning (low intensity aerobic group, resistance balance group, functional group). Real-time feedback conditions: when heart rate variability HRV is lower than 0.3; Decision-making recommendations: It is recommended to increase recovery time and perform deep breathing relaxation training; Step 4.8 Hypoxemia warning (low-intensity aerobic group, resistance balance group, functional group), Real-time feedback conditions: when blood oxygen saturation is lower than 90% and health index is less than 0.5; Decision-making recommendations: It is recommended to reduce the intensity of aerobic exercise and perform intermittent breathing training; Step 4.9 Overtraining warning (low-intensity aerobic group, resistance balance group), Real-time feedback conditions: when the activity time exceeds 70% and the rest time is less than 20%; Decision-making recommendations: It is recommended to increase rest time to avoid overtraining; Step 4.10 Insufficient Resistance Training Load Warning (Balanced Resistance Group) Real-time feedback conditions: When the electromyographic signal shows that the muscle activity level is too low and the health index is greater than 0.5; Decision-making suggestion: It is recommended to appropriately increase the resistance training load to 25% of the maximum load; Step 4.11 High Heart Rate Warning (Low-Intensity Aerobic Group) Real-time feedback conditions: When the heart rate continuously exceeds 85% of the maximum heart rate and the health index is less than 0.5; Decision suggestion: It is recommended to immediately reduce the training intensity and rest for 5 minutes; Step 4.12 Hyperthermia Warning (Low-Intensity Aerobic Group, Resistance Balance Group, Functional Group), Real-time feedback conditions: when body temperature is too high and blood oxygen saturation is lower than 95%; Decision suggestion: Stop training and cool down to avoid overheating; Step 4.13 Muscle fatigue warning, training reduction (low-intensity aerobic group, resistance balance group), Real-time feedback conditions: When the electromyographic signal continuously exceeds the fatigue threshold and the health index is lower than 0.3; Decision-making recommendations: It is recommended to reduce the number of training sessions and monitor muscle fatigue; Step 4.14 Balance Training Suggestions (Resistance Balance Group) Real-time feedback conditions: when the health index is greater than 0.5 and no fall event occurs; Decision-making suggestion: It is recommended to try moderate-intensity balance training and single-leg standing training; Step 4.15 Training intensity adjustment suggestions (low-intensity aerobic group, resistance balance group, functional group), Real-time feedback conditions: when heart rate variability is high and health index is greater than 0.7; Decision-making suggestion: It is recommended to appropriately increase the difficulty of training, increase resistance and intensity; Step 4.16 Gait instability warning (resistance balance group, functional group), Real-time feedback condition: when the gait frequency variance exceeds 0.3; Decision-making suggestion: It is recommended to conduct special balance training and stand on one leg to enhance posture control ability; Step 4.17 Blood oxygen decreasing trend warning (functional group), Real-time feedback conditions: When blood oxygen saturation continues to decrease and its change slope is less than -0.01; Decision Recommendation: It is recommended to check the patient's breathing pattern and continuously monitor lung function.

Images